Skip to content

AI NewsPublished 6 min read

Wiz Ties Black October Breaches to AI Key Abuse

data streams converging into a single bright signal
Listen to this article · 9:14 · AI-generated narration
0:00 / 9:14
Chapters

The short version

Wiz said a single month brought a string of breach disclosures at services that host AI applications, with attackers hunting the stored keys customers use to reach AI models. Northflank customers have reported unauthorized use of AI keys kept on that platform, Bleepingcomputer reported, and the company told them to treat every stored secret as stolen.

  • Fireworks, Zeabur and Lightning traced their incidents to credential access, according to Wiz.
  • Zeabur is shutting down after its incident.
  • Wiz saw stolen enterprise keys first misused within 30 minutes of a vendor breach.
  • Northflank has not said how attackers got in or how many customers were affected, Bleepingcomputer reported.

Wiz tallies a month of breaches

Cloud security company Wiz said in an October 30, 2026, post on the Wiz Blog, titled "Black October" for AI Developer Platforms, that a series of agentic attacks on services hosting AI applications has been leading to LLMJacking at scale. Pasqualepillitteri reported on October 31, 2026, that Wiz counted at least nine AI developer platforms disclosing an incident in October. The list runs from Fireworks AI on October 6 to Northflank on October 30, according to Pasqualepillitteri.

In its post, Wiz wrote: "Any other month, and this would be the top security story." With so many incidents under way, few people connected one attack to the next, Pasqualepillitteri reported, naming the Cloud Security Alliance as the exception Wiz cites. Most of the incidents are still being analyzed, and detailed technical information is not available, Wiz said.

Attackers hunted stored AI keys

Fireworks, Zeabur and Lightning all attribute their incident root causes to credential access, Wiz said. Lightning was hit by a supply chain attack within the Mini-Shai Hulud campaign, according to Wiz. All public reports highlight targeted exploitation that explicitly hunts stored environment variables, with a focus on those granting AI access, the company wrote.

Pasqualepillitteri described environment variables as the fields where a hosting service stores the access keys to AI models. Railway's preliminary report described an attacker hunting for AI provider keys to carry out distillation, meaning copying a model's capabilities by querying it at scale, Pasqualepillitteri reported. Zeabur is shutting down because of its incident, according to both Wiz and Pasqualepillitteri.

Wiz said no single vulnerability or dependency links the attacks. The company tied the picture to two trends it has tracked: agentic attacks, which it said let threat actors compromise victims faster, and LLMJacking, backed by LLM brokers that make stolen credentials easily monetizable. Demand for distillation, notably alleged against Chinese AI labs, drives the market as well, Wiz added.

Northflank says customers' AI keys were abused

Northflank, a cloud application hosting company, disclosed a security incident on October 30 after receiving a threat alert from Wiz, Bleepingcomputer reported on October 31, 2026. "Assume all environment variables have been decrypted. Disable/rotate all keys," Northflank warned in an October 30 update, Bleepingcomputer said. The company later confirmed the abuse, saying, "Users have reported unauthorised use of AI API keys supplied to the Northflank platform."

Wiz warned Northflank at 11:43 UTC on October 30, and engineers reviewed the alert one minute later, according to the incident timeline Northflank published. A network policy patch came first, with more patches at 12:50 UTC and 13:05 UTC, and Northflank asked Wiz's incident response team for help, Bleepingcomputer reported.

Northflank enabled credential rotation for PostgreSQL, MongoDB, MySQL and Redis, and turned on audit logs for all customers, including free accounts, Bleepingcomputer reported. Customers were also told to review AI provider usage and billing logs and report suspicious activity to Northflank and their AI providers, according to Bleepingcomputer. Northflank has not disclosed how attackers gained access, how many customers were affected, or whether credentials beyond AI API keys were stolen or abused, the outlet said.

Stolen keys were misused within minutes

Wiz said it observed downstream abuse of enterprise credentials whose compromise it could attribute to one of the vendor breaches. In a pair of cases, the first unauthorized activity came within 30 minutes of the reported credential access at the breached vendor, Wiz wrote. The first call was a single InvokeModel request from a residential IP address, part of a one-minute sweep that tested stolen keys across multiple organizations, according to Wiz.

Pasqualepillitteri reported that unauthorized charges reached $1,700 for some users in one of the October incidents it covered; the figure has not been confirmed elsewhere.

In a broader Black October feature that also covered weaknesses in widely used website software, Fox News reported on October 30, 2026, that startup founder Colin Parsons woke up to roughly 50 billing notifications and $449 in charges he did not recognize. Fox News called the episode "an expensive surprise that highlights how quickly stolen credentials can become a financial headache."

Fortune tracks a wider wave of incidents

Fortune reported on October 29, 2026, that AI developers, government agencies and private companies disclosed a wave of cybersecurity incidents over several weeks. Cybersecurity experts are calling the month "Black October," Fortune wrote. Some incidents involved hackers using AI tools to hammer defenses at machine speed, while others resulted from plain human error, the magazine reported.

"AI is making the old vulnerabilities cheaper to discover and exploit, but on the other side, the defense side… that's not sped up," Junfeng Yang, a computer science professor at Columbia University, told Fortune. Yang and other experts believe the asymmetry could be temporary, Fortune reported.

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) added flaws known collectively as "php2shell" to its catalog of actively exploited vulnerabilities and ordered civilian federal agencies to fix affected systems by October 29, Fortune reported; the report has not been confirmed elsewhere. Five days earlier, CISA had said it had not yet seen exploitation, according to Fortune.

Tron's take

My take: the October incidents put one concrete question in front of any business running AI features on a third-party hosting service, namely where its AI keys live and who could spend against them. Northflank told customers to assume every environment variable had been decrypted. I would treat that as the default posture for any key stored with a vendor on Wiz's list: rotate it, then check AI provider billing for charges nobody approved.

None of this argues for freezing AI work, and it does not support the opposite view that every new release must be adopted the week it ships. My advice is the deliberate path: keep the AI tools that already proved themselves, keep keys in as few places as possible, and ask each AI provider whether it offers hard spending limits and usage alerts. Owners who call AI news irrelevant to a small business have a fair point about most launch headlines, but stored keys being tested within minutes of a vendor breach deserves an hour of an IT lead's time. My reading is that scattered credentials, more than any single vendor, are the exposure; that is my reading of the news, not a reported result. Earlier XL.net coverage of token resets after the Hugging Face agent incident covers related ground.

XL.net sells security assessments and incident response, including the credential inventory and key rotation work that Northflank's warning points to.

Questions I'd expect

Which companies were named in the October incidents?

Wiz and Pasqualepillitteri named Fireworks AI, Zeabur, Lightning, Railway and Northflank among the affected services.

What did Northflank ask customers to do?

Disable or rotate all keys, treat environment variables as decrypted, and review AI provider usage and billing logs for unexpected activity, Bleepingcomputer reported.

Why do attackers want stolen AI keys?

Wiz said LLM brokers create a market that makes the theft easily monetizable and that demand for model distillation adds to it. Railway's preliminary report described an attacker seeking AI provider keys for distillation, Pasqualepillitteri reported.

All AI news